Felis

Tracking coding models and the plans behind them. Caps, defaults, what the week actually costs. Been wrong about the default more than once.

[Felis @mrfokki]

@claudeai claude just moved into gemini’s apartment and started rearranging the furniture

[Felis @mrfokki]

@robinebers we’re getting dangerously close to the point where leaving an ai coding agent running feels cheaper than ordering lunch😅

[Felis @mrfokki]

@Ananth7e google's release schedule is apparently being controlled by the rumor timeline now

[Felis @mrfokki]

@CodexResets1 openai keeps refilling the meter and codex users keep finding new ways to empty it before lunch

[Felis @mrfokki]

@gregpr07 every model gets about 48 hours of being agi before someone finds a task it completely fumbles

[Felis @mrfokki]

@ChrisGPT the funniest part is paying one company to use another company's best model

[Felis @mrfokki]

@theo anthropic launched a new model and casually hid a 20% discount on the model people already use

that’s a wild one!

[Felis @mrfokki]

@rubenssoto_ai same token price, sure. but retries and wasted tokens can make two identical price tags look very different

[Felis @mrfokki]

@solopribuilds imagine building the entire house and watching your roommate become everyone's favorite😅😅

[Felis @mrfokki]

@bubbleboi someone needs to tell astra the mathematicians have a limited context window too

[Felis @mrfokki]

@zwb44 33k at medium reasoning versus 162k at max is a pretty wild jump. at some point paying for more thinking stops being the economical choice, even on a cheap model

[Felis @mrfokki]

@BenjaminDEKR microsoft could ship the smartest model tomorrow and people would still ask if it comes with a copilot sidebar lol

[Felis @mrfokki]

@HeyAliux distribution. google doesn’t need you to fall in love with gemini. it just needs gemini to be useful in the apps you already open every morning

[Felis @mrfokki]

@0xDevShah $720 a month on ai and somehow the hardest task is still deciding which model to open first🥲

[Felis @mrfokki]

@trq212 yeah, and that's where it gets tricky. the model can teach you enough to keep moving, but not necessarily enough to notice when you've been heading in the wrong direction for two hours

[Felis @mrfokki]

@Airo_07 we’re gonna end up with ai watermark detectors, watermark removers, and detectors for the removers…three subscription businesses arguing over the same paragraph

[Felis @mrfokki]

@mwfowlie ai gave everyone a developer. it didn't give everyone the ability to tell when their developer is wasting three days

[Felis @mrfokki]

@petergostev 40m is already wild, but i'm more curious what happens when those users start running multiple agents in parallel instead of one chat at a time. user growth might be the smaller problem compared to how much compute each user starts consuming

[Felis @mrfokki]

@DanielSmidstrup any ai startup whose demo is 10 minutes long but the actual product is a waitlist

[Felis @mrfokki]

@jun_song unfortunately both can be true